FitMyLLM
NVIDIA· HOPPER

NVIDIA H100 PCIe 96 GB

Running LLMs on the H100 PCIe 96 GB — the long read: which models fit at which quantisation, and the settings worth changing. · Or what else $25,000 buys

VRAM
96 GB
FLAGSHIP
BANDWIDTH
3360
GB/S
MODELS Q4
392/449
87%
7B Q4 SPEED
~384
BLAZING
▸ MODEL COVERAGE @ Q487% OF ALL
▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

Average speeds at Q4 quantization. Actual performance varies by model architecture and context length.

3B
~896
TOK/S
7B
~384
TOK/S
14B
~192
TOK/S
32B
~84
TOK/S
70B
~38
TOK/S
▸ MEASURED RIG REPORTS

We rent the machine and time every model on it: decode, VRAM peak, concurrency, watts and cost per million tokens. Yours may already be one of them — and two are free to read in full.

SEE THE REPORTS
▸ SPECIFICATIONS
VRAM
96 GB
BANDWIDTH
3360 GB/s
FP16 COMPUTE
248.3 TFLOPS
TDP
700W
MEMORY
HBM3
ARCHITECTURE
Hopper
CUDA CORES
16,896
TENSOR CORES
528
PCIE
Gen 5 x16
MSRP
$25,000
384
FAST MODELS · >30 TOK/S
Real-time chat speed
392
USABLE · >10 TOK/S
Comfortable for all tasks
392
TOTAL COMPATIBLE
Fit in VRAM at Q4
▸ RENT IT IN THE CLOUD

Buying H100 PCIe 96 GB costs $15–$40k and isn’t practical for most teams. Spin one up by the hour instead:

Spin up in ~60s. Pay by the second. Cancel anytime.

Some links are affiliate links — we may earn a small commission at no extra cost to you. This helps keep FitMyLLM free and independent.

▸ COMPATIBLE MODELS· 392
S
Falcon-H1R Tiny 90M0.09B
FALCON·256K CTX· CHAT· REASONING
29867
TOK/S · 1% VRAM
S
SmolLM2 135M0.135B
SMOLLM·2K CTX· CHAT
19911
TOK/S · 1% VRAM
S
nomic-embed-text-v1.5 100M0.14B
EMBEDDING·8K CTX· CHAT
19200
TOK/S · 1% VRAM
S
GPT-2 124M0.14B
GPT2·1K CTX· CHAT
19200
TOK/S · 1% VRAM
S
SmolVLM 256M0.256B
SMOLLM·8K CTX· CHAT· VISION
10500
TOK/S · 1% VRAM
S
Gemma 3 270M0.27B
GEMMA·32K CTX· CHAT
9956
TOK/S · 1% VRAM
S
Snowflake Arctic Embed M v2.00.305B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
8813
TOK/S · 1% VRAM
S
bge-large-en-v1.5 335M0.335B
EMBEDDING·1K CTX· CHAT
8024
TOK/S · 1% VRAM
S
mxbai-embed-large-v10.335B
EMBEDDING·1K CTX· EMBEDDING
8024
TOK/S · 1% VRAM
S
Snowflake Arctic Embed L0.335B
EMBEDDING·1K CTX· EMBEDDING
8024
TOK/S · 1% VRAM
S
LFM2 350M0.35B
LFM·125K CTX· CHAT· TOOL_USE
7680
TOK/S · 1% VRAM
S
SmolLM2 360M0.36B
SMOLLM·8K CTX· CHAT
7467
TOK/S · 1% VRAM
S
GPT-2 Medium 345M0.38B
GPT2·1K CTX· CHAT
7074
TOK/S · 1% VRAM
S
Qwen 2.5 0.5B0.5B
QWEN·32K CTX· CHAT
5376
TOK/S · 1% VRAM
S
SmolVLM 500M0.5B
SMOLLM·8K CTX· CHAT· VISION
5376
TOK/S · 1% VRAM
S
Falcon-H1 0.5B0.52B
FALCON·128K CTX· CHAT
5169
TOK/S · 1% VRAM
S
BGE-M30.568B
EMBEDDING·8K CTX· EMBEDDING
4732
TOK/S · 1% VRAM
S
Snowflake Arctic Embed L v2.00.568B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
4732
TOK/S · 1% VRAM
S
Qwen3-Embedding 0.6B0.6B
EMBEDDING·32K CTX· EMBEDDING· MULTILINGUAL
4480
TOK/S · 1% VRAM
S
Falcon-H1R Tiny 0.6B0.6B
FALCON·32K CTX· CHAT· REASONING
4480
TOK/S · 1% VRAM
S
Falcon Perception 0.6B0.6B
FALCON·4K CTX· VISION
4480
TOK/S · 1% VRAM
S
Qwen 1.5 0.5B0.62B
QWEN·32K CTX· CHAT
4335
TOK/S · 1% VRAM
S
LFM2 700M0.74B
LFM·125K CTX· CHAT· TOOL_USE
3632
TOK/S · 1% VRAM
S
Qwen3 0.6B0.75B
QWEN·32K CTX· CHAT· REASONING
3584
TOK/S · 1% VRAM
S
GPT-2 Large 774M0.81B
GPT2·1K CTX· CHAT
3319
TOK/S · 1% VRAM
S
Qwen 3.5 0.8B0.87B
QWEN·256K CTX· CHAT· CODING· MULTILINGUAL
3090
TOK/S · 1% VRAM
S
InternVL3 1B0.94B
OTHER·32K CTX· CHAT· VISION
2860
TOK/S · 1% VRAM
S
MiniCPM5 1B1.08B
MINICPM·128K CTX· CHAT· REASONING· MULTILINGUAL
2489
TOK/S · 1% VRAM
S
TinyLlama 1.1B1.1B
LLAMA·2K CTX· CHAT
2444
TOK/S · 1% VRAM
S
LFM2.5-1.2B-Thinking1.2B
LFM·122K CTX· CHAT· REASONING· TOOL_USE
2240
TOK/S · 1% VRAM
S
Llama-3.2-1B1.2B
LLAMA·4K CTX· CHAT
2240
TOK/S · 1% VRAM
S
LFM2 1.2B1.2B
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
2240
TOK/S · 1% VRAM
S
Zamba2 1.2B1.2B
OTHER·4K CTX· CHAT
2240
TOK/S · 1% VRAM
S
EXAONE-4.0-1.2B1.3B
EXAONE·64K CTX· CHAT
2068
TOK/S · 1% VRAM
S
OPT 1.3B1.3B
OPT·2K CTX· CHAT
2068
TOK/S · 1% VRAM
S
MiniCPM-V 4.61.3B
OTHER·256K CTX· CHAT· VISION
2068
TOK/S · 1% VRAM
S
DeepSeek Coder 1.3B1.35B
DEEPSEEK·16K CTX· CODING
1991
TOK/S · 1% VRAM
S
Phi-1 1.3B1.42B
PHI·2K CTX· CODING
1893
TOK/S · 1% VRAM
S
Phi-1.5 1.3B1.42B
PHI·2K CTX· CHAT· CODING
1893
TOK/S · 1% VRAM
S
LFM2 8B A1B8.3BMoE
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
1792
TOK/S · 6% VRAM
▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your H100 PCIe 96 GB.

WHAT THIS CARD IS WORTH

H100 PCIe 96 GB holds 392 of the models in our catalogue and is, in practice, a IQ4_XS card — the largest it takes is dots.llm1.inst 142.8B at IQ4_XS.

TOKENS/SEC PER $100
0.1
8B at Q4_K_M, so cards compare like for like
VRAM PER $100
0.38 GB
what memory costs on this card
THE BIGGEST IT TAKES
dots.llm1.inst 142.8B142.8B · IQ4_XS86.0 GB2 tok/sEST
WizardLM 2 8x22B141B · Q4_K_S85.7 GB6 tok/sEST
Mixtral-8x22B140.6B · Q4_K_S85.5 GB6 tok/sEST
DBRX 132B132B · Q4_K_M83.8 GB6 tok/sEST
Mistral Medium 3.5128B · Q4_K_M82.1 GB2 tok/sEST
Pixtral Large 124B124B · Q4_K_M79.7 GB2 tok/sEST
Nemotron 3 Super 120B-A12B123.61B · Q4_K_M78.4 GB19 tok/sEST
Mistral-Large 123B123B · Q4_K_M79.1 GB2 tok/sEST
AGAINST CARDS OF SIMILAR MEMORY
Radeon Instinct MI300128 GB$12,0005.8 tok/s per $100
Radeon Instinct MI250128 GB$12,0002.9 tok/s per $100
Instinct MI250X128 GB$10,0003.5 tok/s per $100
M1 Ultra (128GB)96 GB$4,9991.4 tok/s per $100
H100 PCIe 96 GB96 GB$25,0000.1 tok/s per $100

Neighbours in memory rather than in price: memory decides whether a card can do the job at all, so two cards of the same size at different prices is the comparison you are making. Ordered by memory, then bandwidth — not by the value column, which is worked out from bandwidth and price alone and therefore rewards a cheap card whatever its software stack does to that bandwidth in practice. Read it as one input, not as a ranking.

▸ DEVICE UNDER TEST

NVIDIA H100 PCIe 96 GB — 96 GB VRAM.

H100 PCIE 96 GB SPEC
BRAND
NVIDIA
VRAM
96 GB HBM3
BANDWIDTH
3360 GB/s
FP16 COMPUTE
248.3 TFLOPS
FP32 COMPUTE
62.1 TFLOPS
CUDA CORES
16,896
TENSOR CORES
528
TDP
700 W
ARCHITECTURE
Hopper
MSRP
$25000
▸ AI CAPABILITY
392/ 449 models @ Q4

With 96 GB VRAM and 3360 GB/s bandwidth, this GPU handles models up to 132B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~384 tok/s.

RENT IT — LIVE PRICES
Checking what the clouds are charging…
§ 01TOP MODELS FOR H100 PCIE 96 GB
392 FIT · SHOWING 20
MODELSIZEVRAM Q4TOK/SAVG
DBRX 132B132B81.2 GB7546.3
Mistral Medium 3.5128B78.7 GB2146.6
Pixtral Large 124B124B76.3 GB2239.3
Nemotron 3 Super 120B-A12B123.61B76.0 GB22453.2
Mistral-Large 123B123B75.7 GB2233.5
Devstral 2 123B123B75.7 GB2238.1
Qwen 3.5 122B A10B122B75.1 GB26956.8
Nemotron 3 Super 120B120B73.8 GB22457.3
Mistral Small 4 119B119B73.2 GB41450.2
GPT-OSS 120B117B72.0 GB52754.1
Command A 111B111B68.3 GB2427.6
GLM 4.5 Air110B67.7 GB22451.0
Qwen 1.5 110B110B67.7 GB2433.4
Llama 4 Scout 17B-16E109B67.1 GB15833.9
Cogito v2 109B MoE109B67.1 GB158
Ling 2.6 Flash107.49B66.2 GB36336.8
Sarvam 105B105B64.7 GB2648.0
Command-R+ 104B104B64.1 GB2652.7
Llama-3.2-90B-Vision-Instruct90B55.5 GB3048.5
Hunyuan A13B80B49.4 GB20781.1